The Structure of DNA in the Nucleosome Core

articles The structure of DNA in the nucleosome core

Timothy J. Richmond & Curt A. Davey

ETH Zu¨rich, Institut fu¨r Molekularbiologie und Biophysik, ETH-Ho¨nggerberg, CH-8093 Zu¨rich The authors contributed equally to this work

......

The 1.9-A˚ -resolution crystal structure of the nucleosome core particle containing 147 DNA base pairs reveals the conformation of nucleosomal DNA with unprecedented accuracy. The DNA structure is remarkably different from that in oligonucleotides and non- histone protein–DNA complexes. The DNA base-pair-step geometry has, overall, twice the curvature necessary to accommodate the DNA superhelical path in the nucleosome. DNA segments bent into the minor groove are either kinked or alternately shifted. The unusual DNA conformational parameters induced by the binding of histone protein have implications for sequence-dependent protein recognition and nucleosome positioning and mobility. Comparison of the 147-base-pair structure with two 146-base-pair structures reveals alterations in DNA twist that are evidently common in bulk chromatin, and which are of probable importance for chromatin ﬁbre formation and chromatin remodelling.

DNA in eukaryotic cells is packaged repetitively into nucleosomes mobility15. Their mechanism of action most probably derives from by means of extensive association with histone proteins1–3. The the innate ability of nucleosomes to ‘slide’ along DNA without hierarchical chromatin structure formed is the genomic substrate releasing it3. relevant to the vital processes of DNA replication, recombination, An accurate, atomic-level description of DNA conformation in transcription, repair and chromosome segregation, and to the the nucleosome core and comparison with naked DNA is essential pathological progression of cancer and viral disease. Although to an understanding of chromatin properties such as those resulting nucleosomal organization of DNA is essentially ubiquitous from nucleosome position and mobility. Previously, the probable throughout genomes and generally repressive to gene expression, sequence-dependent features of nucleosomal DNA were proposed it also contributes to gene transcription in a gene-specific manner, by extrapolation from high-resolution DNA oligonucleotide struc- suggesting that nucleosome positioning in gene promoter regions is tures16. Here, a direct analysis of nucleosome DNA is made, important for genuine gene regulation in vivo4–6. The question permitted by the high quality of the DNA structure in the 1.9-A˚ - therefore arises of how chromatin structure, in which DNA is resolution X-ray structure of the nucleosome core particle normally highly compacted, permits site-specific access to regula- (NCP147)17,18, comparable to high-resolution oligonucleotide tory factors and more extensive exposure to the transcription structures. The persistently tight curvature of nucleosome core apparatus. DNA causes it to take on conformations not obviously present in The answer is likely to require a knowledge of DNA conformation oligonucleotides but about which there has been speculation for in the nucleosome core. The core comprises 147 base pairs (bp) of many years19. Furthermore, comparison of the DNA twist values DNA and the histone octamer; compared with the nucleosome, it measured for the 147-bp and 146-bp structures with the DNA lacks only 10–90 bp of linker DNA envisaged to be naked or bound sequence periodicity found in bulk chromatin indicates that the to histone H1. The histone-fold domains of the octamer organize DNA stretching observed in the nucleosome core is commonplace the central 129 of 147 bp in 1.59 left-handed superhelical turns with in vivo. a diameter only fourfold that of the double helix. The relatively straight 9-bp terminal segments contribute little to the curvature of Excess DNA curvature the complete 1.67-turn superhelix7. So far, the site-specific regulat- The ideal DNA superhelix that best fits the NCP147 DNA was ory factors that have been discovered bind the linker or terminal constructed from uniformly distributed base pairs by using B-form regions of the intact nucleosome. The lack of binding to the central DNA geometry. It has a radius of 41.9 A˚ and a pitch of 25.9 A˚ region of the superhelix might simply be a consequence of bending (Fig. 1a). Each base-pair step comprising two adjacent base pairs the double helix or, additionally, of unusual DNA conformations contributes 4.538 to the curvature of the ideal superhelix. However, induced by histone binding. At least one protein, HIV-1 integrase, the NCP147 superhelix is not bent uniformly owing to (1) the does prefer DNA bent around the nucleosome in contrast to naked anisotropic flexibility of DNA, (2) local structural features intrinsic DNA8. to the DNA sequence, and (3) irregularities dictated by the under- The initiation of DNA-dependent nuclear processes in the con- lying histone octamer. The NCP147 DNA is generally B-form as text of chromatin implies that nucleosome position is biased by the judged by the criterion of the phosphate ‘Z-coordinate’ with a value DNA sequence to facilitate access by initiation factors. Numerous of 20.18 ^ 0.69 A˚ (mean ^ s.d.; range 22.32 to 1.45 A˚ ). On the examples of positioned nucleosomes in gene promoter regions have basis of oligonucleotide structures, values falling in the range from been described both in vivo and in vitro9–11. Preferential positioning 21 to 0 are indicative of the B-form, whereas values above 2 signify could place factor-binding sequences in nucleosome linker or the A-form20. Other standard conformational parameters also terminal region DNA. Furthermore, nucleosomes are intrinsically indicate that the DNA is predominantly B-form (see Supplementary mobile and yield access to their DNA in vitro, allowing even RNA Table 1). polymerase to transcribe nucleosomal DNA without causing dis- Remarkably, the NCP147 DNA has double the base-pair-step sociation of the histone octamer12–14. In vivo, energy-dependent curvature necessary to produce the DNA superhelix path. The ideal chromatin remodelling factors, targeted by gene regulatory proteins superhelix fit to the NCP147 DNA yields 1.67 superhelical turns for and acting directly on the nucleosome core, augment nucleosome 133.6 bp (1.84 superhelical turns for 147 bp), bending the DNA

NATURE | VOL 423 | 8 MAY 2003 | www.nature.com/nature © 2003 Nature Publishing Group 145 articles through 600.08, whereas the actual curvature of NCP147 DNA is for the ideal and actual curvatures, respectively, and thus the 1,333.38 (both scalar curvature values were calculated with the terminal regions do not account for the discrepancy. The excess is program Curves21). The use of only the central 129 bp to remove not adequately explained by the zigzag path of the NCP147 super- the effect of straightening of the superhelical path at the termini helix, as the overall root-mean-square deviation from the ideal path yields values of 579.38 and 1248.68 (9.75 ^ 5.138 per base-pair step) is only 1.42 A˚ (Fig. 1a). We find that the greater part of the excess curvature is a consequence of alternation of DNA form not revealed by criteria based on oligonucleotide structures. Determining the component of curvature directed towards superhelix formation exposes the origin of the form changes. The potential contributions to curvature from the roll and tilt (see Supplementary Fig. 1) of each base-pair step vary as sinusoidal functions of the accumulated twist angle (V) along the length of the double helix. To evaluate the actual contributions, the roll and tilt values for each base-pair step are multiplied by the corresponding cos V (CAT) and sin V (SAT), respectively. A convenient origin for the summation of V is taken as the central base pair located on the pseudo-two-fold symmetry axis of the particle (base pair 0 or dyad position). For the ideal superhelix, the CAT and SAT values multiplied by 4.538 yield the roll and tilt contributions to superhelix curvature at each base-pair step (iCATand iSAT; Fig. 1b). The actual roll and tilt values for NCP147 DNA are moderately well correlated with CAT (R ¼ 0.72, P , 0.0001) and SAT (R ¼ 0.54, P , 0.0001), respectively. The sum of roll and tilt multiplied by CAT and SAT, respectively, over all base-pair steps yields a curvature free of the effect of path deviations between the ideal and actual superhelices. The resulting value is 897.08, or 1.5-fold that required to accommodate the full NCP147 superhelix. Unlike the computer-generated ideal superhelix, for which curvature derives equally from roll and tilt (iCAT and iSAT; Fig. 1b), curvature for oligonucleotide and NCP147 DNA stems predomi- nately from the roll parameter. For DNA oligonucleotide structures, roll is favoured over tilt by 2:1 (ref. 22), presumably because it requires less base-pair unstacking for any particular bend angle. In the nucleosome core, this ratio is virtually identical at 1.9:1, but when the components that contribute to the superhelix only are considered, the ratio is 3.0:1. Curvature into the superhelix is therefore accumulated every approximately five base pairs, where either the major or minor groove faces the histone octamer. Never- theless, the NCP147 tilt component contributes 77% of that for the ideal superhelix (Fig. 1b; compare tilt and iSAT plots). In contrast, the roll component is 222% of the ideal contribution (Fig. 1b; compare roll and iCAT plots). Analyses of crystal structures of DNA oligonucleotides and protein–DNA complexes normally equate DNA bending with DNA curvature23 and rely on base-pair-step parameters to explain bending behaviour; for example, by using roll and tilt22,24, roll and twist16,25,26 or roll, tilt, twist, slide, shift and rise24,27. This type of approach is inadequate for nucleosome core DNA, given the substantial excess curvature that contributes to superhelix formation. Further analysis indicates that the conformation of nucleosome core DNA is not typical of oligonucleotide DNA and has not previously been described in protein–DNA complexes. Principal- component analysis of DNA structural parameters of both nucleosome core and oligonucleotide DNA show that the two most significant base-pair-step parameter correlations are roll–slide– Figure 1 Superhelical path and base-pair-step parameters for NCP147 DNA. twist and tilt–shift (see Supplementary Table 2). However, the a, Structural alignment of the NCP147 (gold) and best-fit, ideal superhelix (red) paths. The coefficients for the base-pair-step components are significantly NCP147 DNA structure is superimposed (gold). The left view is down the superhelix axis larger for NCP147 DNA (mean magnitudes are 0.88 versus 0.57 and the right view is rotated 908 around the pseudo-two-fold axis (vertical). b, Roll, tilt, for roll–slide–twist and 0.86 versus 0.69 for tilt–shift), indicating shift, slide and twist base-pair-step parameters. The base-pair-step values plotted are the that the NCP147 DNA is limited to a smaller regime of confor- means for the two halves of the symmetrical sequence (shown above, with the dyad mational space than is occupied by oligonucleotide DNA. When the position labelled ‘0’). The iCAT and iSAT curves (red) show the roll and tilt contributions to backbone angles are also included in the analysis, the 1, z and b the ideal superhelix curvature. The minor-groove blocks show base-pair-step shift angles are found tightly linked to the roll–slide–twist component. alternation (pink in shift and tilt) or kinks (orange in roll, slide and twist). The primary Stronger parameter coupling occurs for blocks of base pairs with the bound-phosphate groups are indicated above the base sequence by pointers (numbered minor groove as opposed to the major groove facing the histone as the 5 0 phosphate of the adjacent base) showing the strand direction (dark, 3 0 ! 5 0 ; octamer (mean magnitudes of 0.69 versus 0.27 for roll–slide–twist light, 5 0 ! 3 0 ) and the interacting histone motif (L1, loop 1; L2, loop 2; A1, a1). with backbone angles included), indicating differences in modes of

DNA bending or in conformational variability for the two H2A–H2B dimer, is a probable determinant of smooth versus orientations. kinked bending. Kinkinginminor-grooveblocksalwaysoccursatasingle Major-groove versus minor-groove curvature CA ¼ TG base-pair step that has a roll angle in the range 2188 to The DNA curvature of the central 129 bp of NCP147 favours the 2278, roughly the total curvature into the superhelix for the entire major groove over the minor groove by 1.3:1 per base-pair step in block (Figs 1b and 2c). These kinked steps have a concomitant large contrast to the approximate 2:1 preference observed for oligo- slide value of over 1.5 A˚ and are also overtwisted to values larger nucleotide DNA (evaluated from data in ref. 20). Minor-groove than 408. They have a conformation that is extreme in roll–slide– bending is facilitated in NCP147 by the insertion of an arginine side twist coupling. Ten of 12 minor-groove blocks have one CA ¼ TG chain into the minor groove at each of the 14 sites at which it faces step and two have none. The six CA ¼ TG steps that are kinked have the histone octamer. The minor groove narrows to a width of CAT values between 20.6 and 21, whereas the four non-kinked 3.0 ^ 0.55 A˚ at these sites18. An insightful analysis is made by CA ¼ TG steps have values more positive than 20.5. As steps with dividing the DNA along its length into blocks of continuous positive more negative CAT values have more potential to contribute to and negative CAT values, defining respectively regions where the superhelix curvature, they are conversely under more stress to kink. major-groove and minor-groove curvatures contribute to super- The occurrence of CA ¼ TG steps in kinks is consistent with their helix formation (Fig. 1b). A single, junction base-pair step joins the properties as gleaned from oligonucleotide structures20: (1) TA and major-groove and minor-groove blocks and cannot contribute CA ¼ TG steps are the most flexible as judged from the standard significantly to superhelix curvature by base-pair-step roll because deviation of roll angles, (2) the mean slide for CA ¼ TG is 1.18 A˚ — these steps have the minimum magnitudes for CAT (20.10 to 0.26). the only base-pair step with a value greater than 1 A˚ (the TA mean The major-groove blocks display ‘smooth’ bending with all roll value is 20.80 A˚ ), and (3) with a mean twist angle of 37.48, angles positive (8.47 ^ 5.188) and with systematic underwinding of CA ¼ TG is one of the most overwound base-pair steps. TA and the DNA indicated by the reduced twist angles and negative slide CA ¼ TG are by far the most flexible of the 10 base-pair steps with displacements (Figs 1b and 2a). The minor-groove blocks exhibit regard to roll, slide and twist, but the propensity of TA for negative ‘smooth’ bending only over the H3–H4 tetramer, with essentially all slide makes it much less likely to kink in the manner observed in base-pair steps having negative roll angles (27.86 ^ 6.028). In NCP147 DNA. Clash of the purines across the minor groove in the contrast, minor-groove blocks are kinked over the H2A–H2B kinks observed is avoided because of large slide values. Further- dimers (25.13 ^ 11.018;Fig.1b).Thisdifferenceinminor- more, TA steps have the largest positive mean roll angle (11.88 versus groove-block bending modes associated with H3–H4 and H2A– the next largest of 6.38 for GG ¼ CC), which is consistent with their H2B is most probably a consequence of specific DNA sequences and appearance only in major-groove blocks of NCP147. The import- not underlying differences in the histone binding or degree of ance of base-pair-step flexibility in nucleosome positioning and curvature (mean values over all base-pair steps: H3–H4, 9.88; stability has been suggested previously29,30. A survey of protein– H2A–H2B, 9.98). Smooth bending in minor-groove blocks is DNA complexes shows that only seven base-pair steps are rolled into associated with large alternation of shift values roughly between the minor groove to a degree comparable to that of NCP147 DNA, 21 and þ1A˚ for four base-pair steps, which relieves steric inter- but none of these is a CA ¼ TG step31. Comparison of the mean roll ference between the base edges (Figs 1b and 2b). This mode of angle for each type of base-pair step within NCP147 DNA and from bending, which is the prime contributor to the observed shift–tilt oligonucleotide structures20 shows that they are only weakly corre- coupling, depends most consistently on the GC base-pair step, and lated with each other (see Supplementary Information). on GG ¼ CC or AG ¼ CT base-pair steps, with one and two in each block, respectively. GC steps have been noted previously for their DNA form and superhelix formation high shift propensity20. The GG ¼ CC and AG ¼ CTsteps are, of all Strikingly for NCP147 DNA, the conformational parameter tip steps, the only two that can form cross-chain hydrogen bonds in the oscillates regularly along the length of the DNA (Fig. 3a). The minor groove as observed for oligonucleotides28 and for NCP147. transitions between negative and positive tip angles represent an The occurrence of two GG ¼ CC/AG ¼ CT base-pair steps in the alternation of the DNA form that manifests itself in the major and minor-groove blocks bound to the H3–H4 tetramer, but not the minor blocks as excess roll. By definition, the excess roll of a series of

Figure 2 DNA bending in the NCP147 DNA. Structures (left) and schematic primary bound-phosphate groups (green), the block-junction phosphate groups (white) representations (right) stress uniformity of curvature in the major-groove blocks (red) (a), and the DNA axes for the NCP147 (gold) and ideal (white) superhelices. The contributions and alternating shift values (b) and localization of curvature in kinks in minor-groove of base-pair-step curvature to superhelix bending are listed with base-pair numbers blocks (c) (yellow for one representative double-helical turn). Also indicated are the (centre).

NATURE | VOL 423 | 8 MAY 2003 | www.nature.com/nature © 2003 Nature Publishing Group 147 articles base-pair steps is the sum of the change in tip angles over the Phosphate–phosphate distances, as opposed to their superhelical corresponding base pairs (Fig. 3b). The source of excess roll in the path component, are not correlated with SAT, curvature or roll NCP147 superhelix is localized primarily to block junctions and the angles, and are only weakly correlated with tip. This is the expected base pairs immediately adjacent on either side, where the tip is behaviour because the phosphodiester backbone is essentially significantly non-zero. The reason for the increased tip-angle incompressible, as described previously for oligonucleotide magnitudes at these sites is that they accommodate the smaller structures32, having in NCP147 a mean phosphate–phosphate radius of curvature on the inside surface of the superhelix compared distance of 6.68 ^ 0.23 A˚ between nearest neighbours along the with the outside, about 32 A˚ compared with 52 A˚ . In principle, this same chain. However, a notable variation in the B-form backbone is radius-of-curvature difference should require shorter phosphate– the anticorrelation of the 1 and z backbone angles giving rise to the 33 phosphate distances (PP) where the DNA backbone faces the BI and BII forms . In NCP147, the population of base-pair steps histones than where it faces away. In fact, it is not the absolute intermediate between BI and BII (that is, BI/II) is larger than for distance that is crucial but the component of the PP distance (PPa) oligonucleotides, but the distribution is not significantly correlated lying parallel to the path of the superhelix. As two adjacent base-pair with other form or backbone parameters. The differences in mean steps were found compensatory for this distance, PPa is obtained phosphate–phosphate distances for BI,BI/II and BII forms separately, from three base pairs (i 2 1toi þ 1), and the direction of the even taking into consideration the location in major-groove and superhelix path is defined by four points on the double-helix axis minor-groove blocks or junctions, are less than 1 s.d. of the mean (i 2 1toi þ 2). Importantly, it is PPa that is shortened on the inside distance overall. Nevertheless, in minor-groove blocks, a BII to BI and lengthened on the outside of the DNA superhelix by the transition occurs at kinks, and BII and BI alternation occurs at sites oscillating tip angles. The difference calculated at each base pair of alternating shift. In the latter case, the BII base pair is displaced between the values of PPa for the two phosphodiester chains (DPPa) into the minor groove and has up to 2408 of propeller twist. This is strongly correlated with tip angle (R ¼ 0.91, P , 0.0001; Fig. 3a). unusual conformation for GG and AG base-pair steps is stabilized The values of DPPa compared with SAT, a direct measure of inside by cross-strand bifurcated hydrogen bonds in the minor groove and outside, are somewhat less well correlated (R ¼ 0.83, between guanosine N2 and pyrimidine O2 atoms. Moreover, the P , 0.0001) owing to DNA-sequence-specific effects such as kink- variability observed between repeating binding sites seems to stem ing in the minor-groove blocks. from a dependence on DNA sequence, not only for the choice of

Figure 3 Oscillation of the base-pair-tip parameter for NCP147 DNA. a, Correlation of phosphate distance that lies in the plane containing the local superhelical path. Rotation of base-pair tip and DPPa, plotted as in Fig. 1b. Excess base-pair-step roll (purple) into the a base pair along its tip axis (red) in the direction indicated decreases PPa on the inside superhelix is a consequence of oscillating base-pair-tip angles (black). DPPa (blue) is the and increases it on the outside of the superhelix. Excess roll for a base-pair step is the difference between PPa values for the two phosphodiester chains at a single base pair. difference between the tip angles for the two contributing base pairs. The DNA blocks b, Coupling between PPa, tip and excess roll. PPa is the component of the phosphate– shown are from SHL0.0 and SHL0.5.

148 © 2003 Nature Publishing Group NATURE | VOL 423 | 8 MAY 2003 | www.nature.com/nature articles smooth or kinked bending in minor-groove blocks but also for the relative to the mean phosphate–phosphate distance of 6.68 A˚ (see precise details of the conformation along the entire DNA. Supplementary Table 3). The values for the phosphate groups are comparable to those for the segments of protein main-chain (0.75 Histone constraints on DNA conformation and 0.46 A˚ ) used for the alignments if the phosphates associated Conformational differences between nucleosome core and oligo- with the structurally deviant H2B are not included. Two phosphate nucleotide DNA are probably important for the recognition, or lack groups separated by 21 bases along each DNA strand are therefore of it, of nucleosome DNA by nuclear factors, and for the folding of well localized on both the H3–H4 and H2A–H2B histone-fold pairs nucleosome arrays into the chromatin fibre. Furthermore, the (for example H4:L1 to H4:L2) through histone interaction. Corre- dependence of nucleosome position and stability on base sequence, spondingly, the DNA segments bridging between the halves of the as demonstrated convincingly by experiments in vitro (for example H3–H4 tetramer and H3–H4 tetramer to H2A–H2B dimer are 10 ref. 11), derives from the energetics of sequence-dependent histone– bases in length. Evidently, the histone proteins impose substantial DNA interactions. Parameterization of an energy function, analo- conformational restraints on the DNA, which responds in a gous to that for protein–DNA complexes34, capable of accurate sequence-dependent manner, such as the smooth and kinked prediction of sequence-dependent behaviour will require many bending modes observed for the minor-groove blocks. The phos- high-resolution structures of the nucleosome core particle contain- phate deviations for the A1 motif are larger than for L1 and L2 ing different DNA sequences. Currently, comparisons between motifs, which might reflect their location between phosphate histones H2A, H2B, H3 and H4 for the three primary types of groups at 10.5 bases. This non-optimal arrangement might decrease DNA-binding motif in the nucleosome core give a preliminary sequence-dependent effects on stability by accommodating either indication of the significance of histone restraints on DNA confor- 10 or 11 bases between A1 motifs and the adjacent L1 and L2 motifs. mational flexibility. A possible example of this multiplicity occurs for H2B A1 at the The principal DNA-binding sites in the nucleosome core particle SHL4.5/5.0 block junction (Fig. 1b). Indeed, the location of A1 are the eight L1L2 loop and four a1a1 helix structures within the relative to L1 and L2 motifs could promote DNA stretching. histone-fold domains7. Although the histone–DNA interactions at each of these sites are complex (a complete tabulation is given in DNA stretching ref. 17), one phosphate group from each DNA strand shows at least The double-helical twist values for NCP147 DNA range from 23.78 one conserved interaction with protein over the four types of core to 47.98 and have a mean value of 34.52 ^ 4.758, or equivalently histone (Figs 1b and 4). These primary bound-phosphate groups 10.43 bp per turn, which corresponds to the laboratory reference are 5 0 to the last base pair for both DNA strands in each minor- frame35. Values from experiments that probe DNA by enzymatic groove block, with one exception at SHL4.5 (SHL: superhelix digestion or hydroxyl radical cleavage correspond to the local location7), resulting from a dissimilar packing arrangement of the reference frame and are calculated from the laboratory frame by a1 and a2 helices for H2B in comparison with the other histones7. removing the geometrical contribution from the superhelix pitch. Separate structural alignments, calculated with the histone main- Assuming a uniform pitch angle a, then TwistðlocalÞ¼ chain segments only, for the L1, L2 and A1 binding motifs (the TwistðlaboratoryÞ 2 ð2psinaÞ=N; where N is the number of base- components of L1L2 loop and a1a1 helix sites) and their associated pair steps in one superhelical turn. Using the values from the ideal phosphate groups show the precision with which the selected superhelix of 25.628 and 78.90 steps for a and N, respectively, phosphate groups are localized by binding to histone (see Sup- NCP147 overall has a local frame twist of 10.30 bp per turn. plementary Table 3). Remarkably, two 146-bp nucleosome core particle structures con- For the L1 and L2 binding motifs, the root-mean-square devi- taining differing DNA sequences have values of 10.23 and 10.15 bp ations for the primary bound-phosphate group (1.27 and 1.39 A˚ ) per turn17. These reduced values are a consequence of DNA and its 3 0 -adjacent neighbour (1.42 and 1.48 A˚ ) are strikingly low stretching by 1–2 bp to satisfy the crystal packing constraints of

Figure 4 Structural alignments of the histone-fold DNA-binding motifs and bound- blue; H4, green; H2A, yellow; H2B, red) with respect to the H3 main-chain and interacting phosphate groups. The motifs are L1 (a), L2 (b) and A1 (c). For each, the main-chain Ca groups. Completely conserved (pink) and partly conserved (cyan) hydrogen bonds and atoms of H4, H2A and H2B were aligned with those for H3. The resulting primary bound- interactions (yellow) with the minor-groove arginine side chain (mR) are shown. 0 phosphate group (Pi ) and 3 -adjacent phosphate group (Piþ1) positions are shown (H3,

7 precision adjacent to the alpha 2 operator in Saccharomyces cerevisiae. EMBO J. 10, 3033–3041 (1991). the DNA termini . The actual 12-bp region stretched is in a different 0 location in the two 146-bp structures, and both regions are distant 11. Flaus, A. & Richmond, T. J. Positioning and stability of nucleosomes on MMTV 3 LTR sequences. J. Mol. Biol. 275, 427–441 (1998). from the DNA termini. Evidently, stretching at the DNA termini 12. Meersseman, G., Pennings, S. & Bradbury, E. M. Mobile nucleosomes—a general behaviour. EMBO J. introduces a twist defect in the double helix that then diffuses to 11, 2951–2959 (1992). predominantly one region in each particle, depending on the DNA 13. Polach, K. J. & Widom, J. Mechanism of protein access to specific DNA sequences in chromatin: A dynamic equilibrium model for gene regulation. J. Mol. Biol. 254, 130–149 (1995). sequence. The distorted region is centred on either an H3–H4 or an 14. Studitsky, V. M., Kassavetis, G. A., Geiduschek, E. P. & Felsenfeld, G. Mechanism of transcription H2A–H2B a1a1 DNA-binding site and is bound at each end by the through the nucleosome by eukaryotic RNA polymerase. Science 278, 1960–1963 (1997). adjacent L1L2 sites. These two 146-bp structures therefore contain 15. Tsukiyama, T. The in vivo functions of ATP-dependent chromatin-remodelling factors. Nature Rev. DNA regions that represent trapped intermediates relevant to a Mol. Cell Biol. 3, 422–429 (2002). 36,37 16. Calladine, C. R. & Drew, H. R. Principles of sequence-dependent flexure of DNA. J. Mol. Biol. 192, twist-defect-diffusion mechanism of nucleosome mobility and 907–918 (1986). twist propagation by chromatin-remodelling enzymes. 17. Davey, C. A., Sargent, D. F., Luger, K., Maeder, A. W. & Richmond, T. J. Solvent mediated interactions DNA isolated from nucleosome core particles prepared from in the structure of the nucleosome core particle at 1.9 A˚ resolution. J. Mol. Biol. 319, 1097–1113 bulk-source endogenous chromatin have a sequence periodicity of (2002). 18. Davey, C. A. & Richmond, T. J. DNA-dependent divalent cation binding in the nucleosome core 35 10.17 bp per turn , which is significantly closer to the local frame particle. Proc. Natl Acad. Sci. USA 99, 11169–11175 (2002). values for the 146-bp versus 147-bp nucleosome core particles. This 19. Crick, F. H. C. & Klug, A. Kinky helix. Nature 255, 530–533 (1975). similarity suggests that nucleosomes in the cell nucleus have evolved 20. El Hassan, M. A. & Calladine, C. R. Conformational characteristics of DNA: Empirical classifications and a hypothesis for the conformational behaviour of dinucleotide steps. Phil. Trans. R. Soc. Lond. A to contain DNA stretched on average by 1–2 bp. The difference 355, 43–100 (1997). between 10.17 bp per turn for nucleosomes and 10.5 bp per turn, 21. Lavery, R. & Sklenar, H. The definition of generalised helicoidal parameters and of axis curvature for accepted for the average twist of naked DNA35, accounts for the irregular nucleic acids. J. Biomol. Struct. Dynam. 6, 63–91 (1988). discrepancy in the observed and effective values (21.67 versus 21.2 22. Young, M. A., Ravishanker, G., Beveridge, D. L. & Berman, H. M. Analysis of local helix bending in crystal structures of DNA oligonucleotides and DNA–protein complexes. Biophys. J. 68, 2454–2468 (ref. 38)) of nucleosome supercoiling. The overtwisting of the DNA (1995). double helix on the nucleosome, due in part to DNA stretching, 23. Olson, W. K. Simulating DNA at low resolution. Curr. Opin. Struct. Biol. 6, 242–256 (1996). resolves the ‘linking number paradox’39 observed for nucleosomes 24. Dickerson, R. E. DNA bending: The prevalence of kinkiness and the virtues of normality. Nucleic Acids assembled on closed circular DNA. Unusual twisting or supercoiling Res. 26, 1906–1926 (1998). 25. Goodsell, D. S. & Dickerson, R. E. Bending and curvature calculations in B-DNA. Nucleic Acids Res. of the linker DNA between nucleosome cores in these arrays is not 22, 5497–5503 (1994). required. Formation of a compact nucleosome higher-order struc- 26. El Hassan, M. A. & Calladine, C. R. Two distinct modes of protein-induced bending in DNA. J. Mol. ture could be facilitated by DNA stretching because it would provide Biol. 282, 331–343 (1998). 27. Packer, M. J. & Hunter, C. A. Sequence–structure relationships in DNA oligomers: A computational a buffer against incompatible DNA linker lengths. Each linker DNA approach. J. Am. Chem. Soc. 123, 7399–7406 (2001). segment running between two nucleosome cores could on average 28. Yanagi, K., Prive, G. G. & Dickerson, R. E. Analysis of local helix geometry in three B-DNA decamers adjust in length by up to four base pairs by stretching or compres- and eight dodecamers. J. Mol. Biol. 217, 201–214 (1991). sing the DNA bound to the neighbouring H2A–H2B and H3–H4 29. Shrader, T. E. & Crothers, D. M. Effects of DNA sequence and histone–histone interactions on nucleosome placement. J. Mol. Biol. 216, 69–84 (1990). histone-fold domains. Not only would this provide for linker length 30. Anselmi, C., Bocchinfuso, G., De Santis, P., Savino, M. & Scipioni, A. Dual role of DNA intrinsic adjustment, it would also alleviate twist angle restrictions of at least curvature and flexibility in determining nucleosome stability. J. Mol. Biol. 286, 1293–1301 (1999). 1408. A 31. Dickerson, R. E. & Chiu, T. K. Helix bending as a factor in protein/DNA recognition. Biopolymers 44, 361–403 (1997). 32. Packer, M. J. & Hunter, C. A. Sequence-dependent DNA structure: The role of the sugar-phosphate Methods backbone. J. Mol. Biol. 280, 407–420 (1998). Crystal preparation, data collection, model refinement and figure preparation were as 33. Fratini, A. V., Kopka, M. L., Drew, H. R. & Dickerson, R. E. Reversible bending and helix geometry in a 18 described previously . The ideal DNA superhelix was calculated by using B-form base- B-DNA dodecamer: CGCGAATTBrCGCG. J. Biol. Chem. 257, 14686–14707 (1982). pair geometry40 and fitted by least squares to the NCP147 DNA (T.J.R., unpublished 34. Olson, W. K., Gorin, A. A., Lu, X. J., Hock, L. M. & Zhurkin, V. B. DNA sequence-dependent observations). The twist offset for the central base pair of NCP147 is ^4.998 (negative for deformability deduced from protein–DNA crystal complexes. Proc. Natl Acad. Sci. USA 95, strand with A central). It is presumably non-zero because of the asymmetry of the AT base 11163–11168 (1998). pair lying on the molecular dyad of the particle. DNA curvature, base-pair and base-pair- 35. Travers, A. A. & Klug, A. The bending of DNA in nucleosomes and its wider implications. Phil. Trans. step parameters, and backbone geometry were calculated with the program Curves21 (see R. Soc. Lond. B 317, 537–561 (1987). Supplementary Methods). Excel (Microsoft) and SigmaPlot (SPSS) were used for the 36. Richmond, T. J. & Widom, J. in Chromatin Structure and Gene Expression (eds Elgin, S. C. R. & analysis of DNA geometry. Correlation coefficients were calculated by using variance Workman, J. L.) 1–23 (Oxford Univ. Press, Oxford, 2000). weighting. Principal components were calculated with IMSL routines (T.J.R., unpublished 37. Widom, J. Role of DNA sequence in nucleosome stability and dynamics. Q. Rev. Biophys. 34, 269–324 observations). (2001). 38. Zivanovic, Y., Goulet, I., Revet, B., Le Bret, M. & Prunell, A. Chromatin reconstitution on small DNA Received 16 January; accepted 12 March 2003; doi:10.1038/nature01595. rings II. DNA supercoiling on the nucleosome. J. Mol. Biol. 200, 267–290 (1988). 1. van Holde, K. E. in Chromatin (ed. Rich, A.) (Springer, New York, 1988). 39. Klug, A. & Lutter, L. C. The helical periodicity of DNA on the nucleosome. Nucleic Acids Res. 9, 2. Kornberg, R. D. & Lorch, Y. Twenty-five years of the nucleosome, fundamental particle of the 4267–4283 (1981). eukaryote chromosome. Cell 98, 285–294 (1999). 40. Arnott, S., Dover, S. D. & Wonacott, A. J. Least-squares refinement of the crystal and molecular 3. Elgin, S. C. R. & Workman, J. L. (eds) Chromatin Structure and Gene Expression (Oxford Univ. Press, structures of DNA and RNA from X-ray data and standard bond lengths and angles. Acta Crystallogr. B Oxford, 2000). 25, 2192–2206 (1969). 4. Simpson, R. T. Nucleosome positioning: Occurrence, mechanisms, and functional consequences. Prog. Nucleic Acid Res. Mol. Biol. 40, 143–184 (1991). Supplementary Information accompanies the paper on www.nature.com/nature. 5. Wolffe, A. P. & Kurumizaka, H. The nucleosome: A powerful regulator of transcription. Prog. Nucleic Acid Res. Mol. Biol. 61, 379–422 (1998). Acknowledgements We thank I. Berger and D. Sargent for comments on the manuscript. This 6. Wyrick, J. J. et al. Chromosomal landscape of nucleosome-dependent gene expression and silencing in study was supported by the Swiss National Science Fund. yeast. Nature 402, 418–421 (1999). 7. Luger, K., Maeder, A. W., Richmond, R. K., Sargent, D. F. & Richmond, T. J. Crystal structure of the nucleosome core particle at 2.8 A˚ resolution. Nature 389, 251–260 (1997). Competing interests statement The authors declare that they have no competing financial 8. Pryciak, P. M. & Varmus, H. E. Nucleosomes, DNA-binding proteins, and DNA sequence modulate interests. retroviral integration target site selection. Cell 69, 769–780 (1992). 9. Kornberg, R. D. & Lorch, Y. Chromatin structure and transcription. Annu. Rev. Cell Biol. 8, 563–587 Correspondence and requests for materials should be addressed to T.J.R. (1992). ([email protected]). The coordinates of the X-ray structures are available from the 10. Shimizu, M., Roth, S. Y., Szent-Gyorgyi, C. & Simpson, R. T. Nucleosomes are positioned with base pair Protein Data Bank as files 1kx3, 1kx4 and 1kx5.